Maximum Margin Distance Learning for Dynamic Texture Recognition

The range space of dynamic textures spans spatiotemporal phenomena that vary along three fundamental dimensions: spatial texture, spatial texture layout, and dynamics. By describing each dimension with appropriate spatial or temporal features and by equipping it with a suitable distance measure, elementary distances (one for each dimension) between dynamic texture sequences can be computed. In this paper, we address the problem of dynamic texture (DT) recognition by learning linear combinations of these elementary distances. By learning weights to these distances, we shed light on how “salient” (in a discriminative manner) each DT dimension is in representing classes of dynamic textures. To do this, we propose an efficient maximum margin distance learning (MMDL) method based on the Pegasos algorithm [1], for both class-independent and class-dependent weight learning. In contrast to popular MMDL methods, which enforce restrictive distance constraints and have a computational complex...

• Type: Conference
• Year: 2010
• Where: ECCV
• Keywords: Computer Vision, Dt Dimensions, Dynamic Textures, ECCV 2010, Spatial Texture